[AMD] ci: reinstall MoRI if Different from Dockerfile-pinned commit during install_dependency

[bingxche]

Summary

• The AMD CI image bakes MoRI at the commit pinned in docker/rocm.Dockerfile (MORI_COMMIT). When a PR bumps MORI_COMMIT ahead of the published image, the image is not rebuilt per-PR, so stage-b-test-large-8-gpu-mi35x-disaggregation-amd runs against a stale MoRI (e.g. missing newer IOEngine APIs such as wait_all).
• Add a step to scripts/ci/amd/amd_ci_install_dependency.sh that reinstalls MoRI inside the running ci_sglang container from the repo/commit pinned in rocm.Dockerfile, mirroring the image's build steps — so the container matches the Dockerfile without rebuilding/republishing the image.
• Guarded on /sgl-workspace/mori, so non-MoRI images (every other suite) skip it.

Test plan

• On the mi35x disaggregation suite, the Install dependencies step logs [MORI] Reinstalling MORI <commit> ... / [MORI] Done., and the disaggregation e2e test runs against the Dockerfile-pinned MoRI.
• Validated together with [PD][MoRI] Drive KV transfers with a sharded synchronous worker pool #26922 (which bumps MORI_COMMIT to d87651c) on branch bingxche/mori-ci-reinstall-26922: https://github.com/sgl-project/sglang/actions/runs/27115280593

Made with Cursor

---

CI States

Latest PR Test (Base): ✅ Run #27122910680
Latest PR Test (Extra): ❌ Run #27122910447

[gemini-code-assist]

Code Review

This pull request updates the AMD CI dependency installation script to dynamically reinstall MORI inside the CI container if the workspace directory exists, ensuring that any updates to the MORI commit in the Dockerfile are reflected in the CI run. The reviewer suggested a safer way to pass host environment variables into the Docker container using docker exec -e and single-quoted strings to prevent shell parsing and escaping issues.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.

[gemini-code-assist]

Passing host variables directly into a double-quoted docker exec bash -c "..." string is fragile and can lead to shell parsing or escaping issues if the variables contain special characters.

Instead, it is much safer and cleaner to pass these variables into the container using docker exec -e and use a single-quoted string for the bash -c command. This avoids host-side expansion and simplifies nested quoting/escaping.

Suggested change

	docker exec ci_sglang bash -c "
	set -euo pipefail
	export MORI_GPU_ARCHS='${MORI_GPU_ARCHS}'
	rm -rf /sgl-workspace/mori
	git clone '${MORI_REPO}' /sgl-workspace/mori
	cd /sgl-workspace/mori
	git checkout '${MORI_COMMIT}'
	git submodule update --init --recursive
	python3 setup.py develop
	python3 -c 'import os, torch; print(os.path.join(os.path.dirname(torch.__file__), \"lib\"))' > /etc/ld.so.conf.d/torch.conf
	ldconfig
	"
	docker exec \
	-e MORI_GPU_ARCHS="${MORI_GPU_ARCHS}" \
	-e MORI_REPO="${MORI_REPO}" \
	-e MORI_COMMIT="${MORI_COMMIT}" \
	ci_sglang bash -c '
	set -euo pipefail
	rm -rf /sgl-workspace/mori
	git clone "$MORI_REPO" /sgl-workspace/mori
	cd /sgl-workspace/mori
	git checkout "$MORI_COMMIT"
	git submodule update --init --recursive
	python3 setup.py develop
	python3 -c "import os, torch; print(os.path.join(os.path.dirname(torch.__file__), \"lib\"))" > /etc/ld.so.conf.d/torch.conf
	ldconfig
	'